Wilcoxon signed rank test - Jasp vs Python, slightly different p values

alebyk

I ran Wilcoxon signed test using scipy.stats.wilcoxon (https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.wilcoxon.html) and Jasp.

I got p=0.005 in python and p=0.002 in Jasp, I was wondering why might that be?

I attach a csv file with the paired data I was testing.

EJ

I'll pass this on to Johnny. We use a particular R package I think.
Cheers,
E.J.

JohnnyB

Hi Alebyk,
The difference lies in whether a normal approximation is used for the p-value. From the SciPy page: "Because the normal approximation is used for the calculations, the samples used should be large. A typical rule is to require that n > 20."
In R (with the wilcox.test function), an exact p-value can be calculated for small sample sizes - since your sample size is only 10, the exact p-value can be calculated (and R, and therefore JASP, does that by default for small n):
wilcox.test(dat$con) # gives p-value = 0.001953
wilcox.test(dat$con, exact = FALSE) # gives p-value = 0.005922

I hope this clarifies things!
Kind regards,
Johnny

DrMark

If it helps - p=.005 in SPSS as well

alebyk

Thanks a lot for clarification! I now see other people have reported this before, e.g. here: https://stackoverflow.com/questions/33579785/difference-between-wilcoxon-test-in-r-and-python

Cheers again, love your software btw!